mirror of
https://gitlab.freedesktop.org/gstreamer/gstreamer.git
synced 2024-12-18 14:26:43 +00:00
videoscale: Added NV support for 4Tap resize
https://bugzilla.gnome.org/show_bug.cgi?id=736845
This commit is contained in:
parent
2133a98eb1
commit
6babe786f7
3 changed files with 141 additions and 1 deletions
|
@ -413,7 +413,7 @@ get_formats_filter (GstVideoScaleMethod method)
|
||||||
"ARGB, BGRA, ABGR, AYUV, ARGB64, AYUV64, "
|
"ARGB, BGRA, ABGR, AYUV, ARGB64, AYUV64, "
|
||||||
"RGB, BGR, v308, YUY2, YVYU, UYVY, "
|
"RGB, BGR, v308, YUY2, YVYU, UYVY, "
|
||||||
"GRAY8, GRAY16_LE, GRAY16_BE, I420, YV12, "
|
"GRAY8, GRAY16_LE, GRAY16_BE, I420, YV12, "
|
||||||
"Y444, Y42B, Y41B, RGB16, RGB15 }");
|
"Y444, Y42B, Y41B, RGB16, RGB15, NV12, NV21 }");
|
||||||
return gst_static_caps_get (&fourtap_filter);
|
return gst_static_caps_get (&fourtap_filter);
|
||||||
}
|
}
|
||||||
case GST_VIDEO_SCALE_LANCZOS:
|
case GST_VIDEO_SCALE_LANCZOS:
|
||||||
|
@ -1392,6 +1392,10 @@ do_scale (GstVideoFilter * filter, VSImage dest[4], VSImage src[4])
|
||||||
vs_image_scale_linear_Y (&dest[0], &src[0], videoscale->tmp_buf);
|
vs_image_scale_linear_Y (&dest[0], &src[0], videoscale->tmp_buf);
|
||||||
vs_image_scale_linear_NV12 (&dest[1], &src[1], videoscale->tmp_buf);
|
vs_image_scale_linear_NV12 (&dest[1], &src[1], videoscale->tmp_buf);
|
||||||
break;
|
break;
|
||||||
|
case GST_VIDEO_SCALE_4TAP:
|
||||||
|
vs_image_scale_4tap_Y (&dest[0], &src[0], videoscale->tmp_buf);
|
||||||
|
vs_image_scale_4tap_NV12 (&dest[1], &src[1], videoscale->tmp_buf);
|
||||||
|
break;
|
||||||
default:
|
default:
|
||||||
goto unknown_mode;
|
goto unknown_mode;
|
||||||
}
|
}
|
||||||
|
|
|
@ -86,6 +86,11 @@ static void vs_scanline_resample_4tap_AYUV64 (uint16_t * dest, uint16_t * src,
|
||||||
static void vs_scanline_merge_4tap_AYUV64 (uint16_t * dest, uint16_t * src1,
|
static void vs_scanline_merge_4tap_AYUV64 (uint16_t * dest, uint16_t * src1,
|
||||||
uint16_t * src2, uint16_t * src3, uint16_t * src4, int n, int acc);
|
uint16_t * src2, uint16_t * src3, uint16_t * src4, int n, int acc);
|
||||||
|
|
||||||
|
static void vs_scanline_merge_4tap_NV (uint8_t * dest, uint8_t * src1,
|
||||||
|
uint8_t * src2, uint8_t * src3, uint8_t * src4, int n, int acc);
|
||||||
|
static void vs_scanline_resample_4tap_NV (uint8_t * dest, uint8_t * src, int n,
|
||||||
|
int src_width, int *xacc, int increment);
|
||||||
|
|
||||||
static double
|
static double
|
||||||
vs_4tap_func (double x)
|
vs_4tap_func (double x)
|
||||||
{
|
{
|
||||||
|
@ -188,6 +193,133 @@ vs_scanline_merge_4tap_Y (uint8_t * dest, uint8_t * src1, uint8_t * src2,
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
|
static void
|
||||||
|
vs_scanline_resample_4tap_NV (uint8_t * dest, uint8_t * src,
|
||||||
|
int n, int src_width, int *xacc, int increment)
|
||||||
|
{
|
||||||
|
int i;
|
||||||
|
int j;
|
||||||
|
int acc;
|
||||||
|
int x;
|
||||||
|
int u, v;
|
||||||
|
int stride = src_width * 2;
|
||||||
|
|
||||||
|
acc = *xacc;
|
||||||
|
for (i = 0; i < n; i++) {
|
||||||
|
j = acc >> 16;
|
||||||
|
j *= 2;
|
||||||
|
x = (acc & 0xff00) >> 8;
|
||||||
|
if (j - 1 >= 0 && j + 2 < src_width) {
|
||||||
|
u = vs_4tap_taps[x][0] * src[j - 2];
|
||||||
|
v = vs_4tap_taps[x][0] * src[j - 1];
|
||||||
|
u += vs_4tap_taps[x][1] * src[j];
|
||||||
|
v += vs_4tap_taps[x][1] * src[j + 1];
|
||||||
|
u += vs_4tap_taps[x][2] * src[j + 2];
|
||||||
|
v += vs_4tap_taps[x][2] * src[j + 3];
|
||||||
|
u += vs_4tap_taps[x][3] * src[j + 4];
|
||||||
|
v += vs_4tap_taps[x][3] * src[j + 5];
|
||||||
|
} else {
|
||||||
|
u = vs_4tap_taps[x][0] * src[CLAMP (j - 2, 0, stride - 1)];
|
||||||
|
v = vs_4tap_taps[x][0] * src[CLAMP (j - 1, 0, stride - 1)];
|
||||||
|
u += vs_4tap_taps[x][1] * src[CLAMP (j, 0, stride - 1)];
|
||||||
|
v += vs_4tap_taps[x][1] * src[CLAMP (j + 1, 0, stride - 1)];
|
||||||
|
u += vs_4tap_taps[x][2] * src[CLAMP (j + 2, 0, stride - 1)];
|
||||||
|
v += vs_4tap_taps[x][2] * src[CLAMP (j + 3, 0, stride - 1)];
|
||||||
|
u += vs_4tap_taps[x][3] * src[CLAMP (j + 4, 0, stride - 1)];
|
||||||
|
v += vs_4tap_taps[x][3] * src[CLAMP (j + 5, 0, stride - 1)];
|
||||||
|
}
|
||||||
|
u += (1 << (SHIFT - 1));
|
||||||
|
v += (1 << (SHIFT - 1));
|
||||||
|
dest[i * 2] = CLAMP (u >> SHIFT, 0, 255);
|
||||||
|
dest[i * 2 + 1] = CLAMP (v >> SHIFT, 0, 255);
|
||||||
|
acc += increment;
|
||||||
|
}
|
||||||
|
*xacc = acc;
|
||||||
|
}
|
||||||
|
|
||||||
|
static void
|
||||||
|
vs_scanline_merge_4tap_NV (uint8_t * dest, uint8_t * src1, uint8_t * src2,
|
||||||
|
uint8_t * src3, uint8_t * src4, int n, int acc)
|
||||||
|
{
|
||||||
|
int i;
|
||||||
|
int val;
|
||||||
|
int a, b, c, d;
|
||||||
|
|
||||||
|
acc = (acc >> 8) & 0xff;
|
||||||
|
a = vs_4tap_taps[acc][0];
|
||||||
|
b = vs_4tap_taps[acc][1];
|
||||||
|
c = vs_4tap_taps[acc][2];
|
||||||
|
d = vs_4tap_taps[acc][3];
|
||||||
|
for (i = 0; i < (n * 2); i++) {
|
||||||
|
val = a * src1[i];
|
||||||
|
val += b * src2[i];
|
||||||
|
val += c * src3[i];
|
||||||
|
val += d * src4[i];
|
||||||
|
val += (1 << (SHIFT - 1));
|
||||||
|
dest[i] = CLAMP (val >> SHIFT, 0, 255);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
void
|
||||||
|
vs_image_scale_4tap_NV12 (const VSImage * dest, const VSImage * src,
|
||||||
|
uint8_t * tmpbuf)
|
||||||
|
{
|
||||||
|
int yacc;
|
||||||
|
int y_increment;
|
||||||
|
int x_increment;
|
||||||
|
int i;
|
||||||
|
int j;
|
||||||
|
int xacc;
|
||||||
|
int k;
|
||||||
|
|
||||||
|
if (dest->height == 1)
|
||||||
|
y_increment = 0;
|
||||||
|
else
|
||||||
|
y_increment = ((src->height - 1) << 16) / (dest->height - 1);
|
||||||
|
|
||||||
|
if (dest->width == 1)
|
||||||
|
x_increment = 0;
|
||||||
|
else
|
||||||
|
x_increment = ((src->width - 1) << 16) / (dest->width - 1);
|
||||||
|
|
||||||
|
k = 0;
|
||||||
|
/* The below loop is to initialize the 4 buffers present in tmpbuf
|
||||||
|
For scaling there are used in cyclic way, so they need to be
|
||||||
|
initialized before entering the main loop */
|
||||||
|
for (i = 0; i < 4; i++) {
|
||||||
|
xacc = 0;
|
||||||
|
vs_scanline_resample_4tap_NV (tmpbuf + i * dest->stride,
|
||||||
|
src->pixels + CLAMP (i, 0, src->height - 1) * src->stride, dest->width,
|
||||||
|
src->width, &xacc, x_increment);
|
||||||
|
}
|
||||||
|
|
||||||
|
yacc = 0;
|
||||||
|
for (i = 0; i < dest->height; i++) {
|
||||||
|
uint8_t *t0, *t1, *t2, *t3;
|
||||||
|
|
||||||
|
j = yacc >> 16;
|
||||||
|
|
||||||
|
while (j > k) {
|
||||||
|
k++;
|
||||||
|
if (k + 3 < src->height) {
|
||||||
|
xacc = 0;
|
||||||
|
vs_scanline_resample_4tap_NV (tmpbuf + ((k + 3) & 3) * dest->stride,
|
||||||
|
src->pixels + (k + 3) * src->stride,
|
||||||
|
dest->width, src->width, &xacc, x_increment);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
t0 = tmpbuf + (CLAMP (j - 1, 0, src->height - 1) & 3) * dest->stride;
|
||||||
|
t1 = tmpbuf + (CLAMP (j, 0, src->height - 1) & 3) * dest->stride;
|
||||||
|
t2 = tmpbuf + (CLAMP (j + 1, 0, src->height - 1) & 3) * dest->stride;
|
||||||
|
t3 = tmpbuf + (CLAMP (j + 2, 0, src->height - 1) & 3) * dest->stride;
|
||||||
|
vs_scanline_merge_4tap_NV (dest->pixels + i * dest->stride,
|
||||||
|
t0, t1, t2, t3, dest->width, yacc & 0xffff);
|
||||||
|
|
||||||
|
yacc += y_increment;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
void
|
void
|
||||||
vs_image_scale_4tap_Y (const VSImage * dest, const VSImage * src,
|
vs_image_scale_4tap_Y (const VSImage * dest, const VSImage * src,
|
||||||
uint8_t * tmpbuf)
|
uint8_t * tmpbuf)
|
||||||
|
|
|
@ -36,6 +36,10 @@ G_GNUC_INTERNAL void vs_image_scale_4tap_Y (const VSImage * dest,
|
||||||
const VSImage * src,
|
const VSImage * src,
|
||||||
uint8_t * tmpbuf);
|
uint8_t * tmpbuf);
|
||||||
|
|
||||||
|
G_GNUC_INTERNAL void vs_image_scale_4tap_NV12 (const VSImage * dest,
|
||||||
|
const VSImage * src,
|
||||||
|
uint8_t * tmpbuf);
|
||||||
|
|
||||||
G_GNUC_INTERNAL void vs_image_scale_4tap_RGBA (const VSImage * dest,
|
G_GNUC_INTERNAL void vs_image_scale_4tap_RGBA (const VSImage * dest,
|
||||||
const VSImage * src,
|
const VSImage * src,
|
||||||
uint8_t * tmpbuf);
|
uint8_t * tmpbuf);
|
||||||
|
|
Loading…
Reference in a new issue